Askmrrobot puts in the disclaimer/clarification: "These weights are optimized for Tier 16 gear and the Legendary Meta Gem. They are also optimal for Tier 15 and near-optimal at lower gear levels." I seems like they ran the numbers at different gear levels and they are confident that their weights are optimized for BiS but more than adequate for lower gear levels
I have never really cared for stat weights, and that's not just because they take forever to run.
They're so heavily dependent on gear, fight conditions, and most importantly (but rarely noted) the priority being tested.
Plus, I fundamentally disagree with the liberal use of extrapolation from a multivariable function (which we don't know but are approximating numerically) by taking the gradient. All of the standard assumptions when extrapolating have to be considered (which are generally fine), but you also have to consider that each variable in the function affects every other variable in that function as well (and this is not necessarily linear). I rarely see these assumptions even considered (mainly when projecting out very far with a set of stat weights).
E.g.:
f( x , y ) = 2 * y * ( x )^2; if you increase x, you have increased the slope on y, which makes increases in y now more pronounced.
df( x , y )/dx = 4 * y * x; increases in x are dependent on y and x
df( x , y )/dy = 2 * ( x )^2; increases in y are strongly dependent on x and not dependent on y
What stat weights do is tell you the highest DPS gain from your current stats, priority, and fight conditions. It's like going down a mountain--stat weights will tell you the steepest slope from your current position, but as soon as you move even an inch, your current position has changed. However, the steepest slope is not necessarily the slope that you are actually on, either (e.g. wrong priority or such heavy influence from stacking a certain stat, like going down the wrong side of the mountain).
Ultimately, I don't really see a way around the issue for guides, though. A stat priority has to be recommended (x > y > z ... is fine for most people, but not for those who want to determine if 500 of x and z is better than 750 of y), but I do think displaying the stat weights for the BiS gear is rather useless. The whole point of stat weights is to look forward (it's a forward difference, after all)--if you're in the BiS gear, you have nothing to gear forward towards. But, I understand that authors can't really assume that their readers have a certain level of gear (like T15H versus T15N going into T16).
Last edited by SSHA778; 2013-09-13 at 11:30 PM.
"I have it all simmed."
Euliat